Research map / Vision-language and multimodal / CLIP and zero-shot
Bridging the Skeleton-Text Modality Gap: Diffusion-Powered Modality Alignment for Zero-shot Skeleton-based Action Recognition
Jeonghyeok Do, Munchurl Kim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add triplet diffusion loss to the diffusion-loss-only model (ablation) | TDSM · NTU-60 55/5 split | top-1 accuracy (%): 79.87 -> 86.49 | Improved |
| add triplet diffusion loss to the diffusion-loss-only model (ablation) | TDSM · NTU-120 96/24 split | top-1 accuracy (%): 57.65 -> 65.06 | Improved |
| add diffusion loss to the TD-loss-only model (ablation) | TDSM · NTU-60 55/5 split | top-1 accuracy (%): 80.90 -> 86.49 | Improved |
| use both global and local text features instead of global only (ablation) | TDSM · NTU-60 55/5 split | top-1 accuracy (%): 83.41 -> 86.49 | Improved |
| use both global and local text features instead of local only (ablation) | TDSM · NTU-120 110/10 split | top-1 accuracy (%): 69.95 -> 74.15 | Improved |
| set total diffusion timesteps T=50 instead of T=1 (ablation) | TDSM · NTU-60 48/12 split | top-1 accuracy (%): 44.10 -> 56.03 | Improved |
| increase total timesteps from 50 to 500 (ablation) | TDSM · NTU-60 55/5 split | top-1 accuracy (%): 86.49 -> 81.34 | Hurt |
| increase total timesteps from 50 to 100 (ablation) | TDSM · NTU-60 48/12 split | top-1 accuracy (%): 56.03 -> 56.27 | No clear effect |
| use random Gaussian noise each training step instead of fixed noise (ablation) | TDSM · NTU-60 55/5 split | top-1 accuracy (%): 76.40 -> 86.49 | Improved |
| use random Gaussian noise each training step instead of fixed noise (ablation) | TDSM · NTU-120 96/24 split | top-1 accuracy (%): 52.21 -> 65.06 | Improved |
| use DiT backbone instead of U-Net (ablation) | TDSM · NTU-60 48/12 split | top-1 accuracy (%): 51.12 -> 56.03 | Improved |
| predict skeleton feature z_x instead of noise (ablation) | TDSM · NTU skeleton ZSAR | noise-induced fluctuation in accuracy: fluctuation reduced by 5x | Mixed |
Closest papers by meaning
- Semantic-guided Cross-Modal Prompt Learning for Skeleton-based Zero-shot Action Recognition (CVPR 2025 · similarity 0.89)
- Part-aware Unified Representation of Language and Skeleton for Zero-shot Action Recognition (CVPR 2024 · similarity 0.88)
- Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton Action Recognition (CVPR 2025 · similarity 0.88)
- Navigating Open Set Scenarios for Skeleton-Based Action Recognition (AAAI 2024 · similarity 0.87)
- SALAD: Skeleton-aware Latent Diffusion for Text-driven Motion Generation and Editing (CVPR 2025 · similarity 0.86)
- Boosting Skeleton-based Zero-Shot Action Recognition with Training-Free Test-Time Adaptation (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
